-
New Photoplethysmographic Signal Analysis
Algorithm for Arterial Stiffness Estimation
Kristjan Pilt, Rain Ferenets, Kalju Meigas, Lars-Göran Lindberg,
Kristina Temitski and
Margus Viigimaa
Linköping University Post Print
N.B.: When citing this work, cite the original article.
Original Publication:
Kristjan Pilt, Rain Ferenets, Kalju Meigas, Lars-Göran Lindberg,
Kristina Temitski and Margus
Viigimaa, New Photoplethysmographic Signal Analysis Algorithm
for Arterial Stiffness
Estimation, 2013, Scientific World Journal, (2013).
http://dx.doi.org/10.1155/2013/169035
Copyright: Hindawi Publishing Corporation
http://www.hindawi.com/
Postprint available at: Linköping University Electronic
Press
http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-97461
http://dx.doi.org/10.1155/2013/169035http://www.hindawi.com/http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-97461http://twitter.com/?status=OA
Article: New Photoplethysmographic Signal Analysis Algorithm for
Arterial Stiffness Estim...
http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-97461 via
@LiU_EPress %23LiU
-
Hindawi Publishing CorporationThe Scientific World JournalVolume
2013, Article ID 169035, 9
pageshttp://dx.doi.org/10.1155/2013/169035
Research ArticleNew Photoplethysmographic Signal Analysis
Algorithm forArterial Stiffness Estimation
Kristjan Pilt,1 Rain Ferenets,1 Kalju Meigas,1 Lars-Göran
Lindberg,2
Kristina Temitski,1 and Margus Viigimaa1
1 Department of Biomedical Engineering, Technomedicum, Tallinn
University of Technology, Ehitajate tee 5, 19086 Tallinn,
Estonia2Department of Biomedical Engineering, Linköping
Univeristy, 581 85 Linköping, Sweden
Correspondence should be addressed to Kristjan Pilt;
[email protected]
Received 10 May 2013; Accepted 1 July 2013
Academic Editors: N. Bouguila, M. Engin, and T. Yamasaki
Copyright © 2013 Kristjan Pilt et al. This is an open access
article distributed under the Creative Commons Attribution
License,which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly
cited.
The ability to identify premature arterial stiffening is of
considerable value in the prevention of cardiovascular diseases.The
“ageingindex” (AGI), which is calculated from the second derivative
photoplethysmographic (SDPPG) waveform, has been used as onemethod
for arterial stiffness estimation and the evaluation of
cardiovascular ageing. In this study, the new SDPPGanalysis
algorithmis proposed with optimal filtering and signal
normalization in time. The filter parameters were optimized in
order to achieve theminimal standard deviation of AGI, which gives
more effective differentiation between the levels of arterial
stiffness. As a result,the optimal low-pass filter edge frequency
of 6Hz and transitionband of 1Hz were found, which facilitates AGI
calculation witha standard deviation of 0.06. The study was carried
out on 21 healthy subjects and 20 diabetes patients. The linear
relationship(𝑟 = 0.91) between each subject’s age and AGI was
found, and a linear model with regression line was constructed. For
diabetespatients, the mean AGI value difference from the proposed
model 𝑦AGI was found to be 0.359. The difference was found
betweenhealthy and diabetes patients groups with significance level
of 𝑃 < 0.0005.
1. Introduction
There has been an increased interest in the development
ofinnovative noninvasive methods and devices for the diagno-sis of
cardiovascular diseases [1–3]. Photoplethysmographic(PPG) waveform
analysis has been used as one method [4].
PPG is a noninvasive optical technique for measuringchanges in
blood circulation that is mainly used for moni-toring blood
perfusion in the skin. The PPG finger sensorconsists of a light
emitting diode (LED), which is often redor infrared, and a
photodetector (PD) [5]. PD and LED areon the opposite side of the
finger. The light is emitted fromthe LED to the skin and a small
fraction of light intensitychanges is received by the PD, which are
related to bloodflow, blood volume, blood vessel wall movement, and
theorientation of red blood cells in the underlying tissue [6,
7].The PPG signal consists of different components: DC andAC
components and noise, which can be caused by thepoor perfusion
state and motion artifacts [5]. Noise can beeliminated by using
different filtering techniques [8].The AC
component is synchronous with the heart rate and dependson
changes in the pulsatile pressure and pulsatile bloodvolume.
It is apparent that the AC component of the PPG signalchanges
with age and the waveform transforms from awavy into a
triangular-shaped signal (Figure 1, upper part).Regarding time
domain, different methods to analyze thewaveform of the PPG signal,
measured at the finger, canbe used for arterial stiffness
estimation and evaluation ofcardiovascular aging [9–11].
One option is to use the second derivative of the PPGsignal
(SDPPG), which was first introduced by Takazawaet al. [12]. The
SDPPG is analyzed by using the amplitudesof the distinctive waves
“a”, “b”, “c”, “d”, and “e”, which aresituated in the systolic
phase of the heart cycle (Figure 1,lower part). The amplitudes of
the waves are normalized asfollows: b/a, c/a, d/a, and e/a. They
found that normalizedamplitude b/a increases and c/a, d/a, and e/a
decrease inproportion to the increase in the subject’s age. As a
result an“ageing index” (AGI) parameter was proposed according
to
-
2 The Scientific World Journal
a
b
cd
e
24-years-old subject’s PPG signal
SDPPG signal
20
0
−20
−40
−60
(a.u
.)
4.5 5 5.5
4.5 5 5.5
Time (s)
Time (s)
0.02
0.01
0
−0.01
−0.02
(a.u
.)
a
b
c
d
e
62-years-old subject’s PPG signal(b)(a)
SDPPG signal(a
.u.)
4.54 5
4.54 5Time (s)
Time (s)
50
0
−50
−100
0.02
0.01
0
−0.01
(a.u
.)
Figure 1:The finger PPG signal and its second derivative with
distinct waves “a”, “b”, “c”, “d”, and “e” of 24-year (a) and
62-year-old (b) subjects.
AGI = (b−c−d−e)/a, where the a, b, c, d, and e are theamplitudes
of the waves. The AGI is used to describe thecardiovascular age of
the subject.
In recent publications, the correlation relationshipbetween
cardiovascular risk factors and the SDPPGnormalized amplitudes
values has been analyzed statistically[13–15]. Normalized
amplitudes of SDPPG and AGI can begood parameters for a screening
method to detect increasesin the stiffness of the arteries
[16].
The sample segment of PPG and SDPPG signal, which hasbeen
registered from a 37-year-old healthy subject, with AGIvalues, is
shown in Figure 2. The SDPPG signal is processed,and the wave
amplitudes are detected according to a studyby Millasseau et al.
[17]. The similar processing method hasbeen also described in less
detail in other studies [12–15]. Itis assumed that the
cardiovascular system does not change
over short periods in cases of healthy subjects. It is
visiblefrom Figure 2 that the AGI values for the healthy subjectare
noticeably higher for the first and third periods. Thedifference
between maximal and minimal AGI values is 0.47,which constitutes
about 39% from the whole scale of AGI[12]. Furthermore, the
detected peaks in the first and thirdperiods are located to the
beginning of systolic phase of thePPG signal compared to the second
and fourth periods. Asa result the amplitudes of detected peaks in
the consecutiveperiods describe different phase of the PPG waveform
andAGI values become noticeably different. This leads to
higherstandard deviation of AGI and to faulty interpretation of
theresults for a single subject. The detection of the peaks
fromdifferent phases of PPG signal in case of consecutive periodsis
due to the insufficient suppression of PPG signal highercomponents
and noise.
-
The Scientific World Journal 3
PPG signal50
0
−50
−100
−150
(a.u
.)
11 11.5 12 12.5 13 13.5 14 14.5Time (s)
11 11.5 12 12.5 13 13.5 14 14.5Time (s)
SDPPG signal
0.04
0.02
0
−0.02
(a.u
.)
a
b
cd
ea
b
cd
ea
b
cd
ea
bcd
e
AGI1 = −0.55 AGI2 = −0.7 AGI3 = −0.37 AGI4 = −0.84
Figure 2: The sample segment of the PPG signal (upper part) from
a 37-year-old healthy subject and its second derivative (lower
part) withdetected wave peaks andAGI values.The SDPPG signal is
processed, and the wave amplitudes are detected according to a
study byMillasseauet al. [17].
The AGI has to be calculated with low standard deviationin order
to differentiate the subjects with increased stiffnessfrom the
healthy subjects. In this study, we have improved theSDPPG analysis
method in order to obtain the AGI valueswith minimal standard
deviation and to detect the waves atthe same locations within one
period of the PPG signal. Thealgorithm is tested on group of
healthy subjects and a smallgroup of diabetes patients as a pilot
study.
2. Methods
2.1. SDPPGAnalysis Algorithm. Normalization of the
period’slength, averaging, filtering, and detection of the waves
fromthe SDPPG signal are illustrated in a block diagram inFigure 3.
The PPG signal is filtered with low- and high-passFIR filters in
order to separate DC components and highfrequency noise.The cutoff
frequencies for the low- and high-pass filters are selected as 30Hz
and 0.5Hz, respectively.Both filters are designed using the window
method, with theHamming window function where the corresponding
filterorders are chosen as 500 for the low-pass after and 4000
forthe high-pass filter.
Subsequently the PPG signal is differentiated two andfour times
(Figure 3). The simplest differentiator calculates
the differences between two consecutive samples of the
signal,which is also known as the first-difference
differentiator.Thiskind of differentiator works as a high-pass
filter, and the highfrequencies are amplified as a result. However,
the unwantednoise is located at higher frequencies for the PPG
signal. Dueto the reason outlined previously, the Smooth Noise
RobustDifferentiator (SNRD) is used.
The SNRD has been developed for different cases thatare
particularly beneficial for carrying out experiments withnoisy data
where differentiation is required [18]. This differ-entiation
scheme possesses the following characteristics: pre-cise at low
frequencies, smooth and guaranteed suppressionof high
frequencies.Theorder of the differentiator determinesthe
suppression of the high frequencies. In this algorithm, thefifth
order of the differentiator is used, which is also the
lowestpossible one. At the lower frequencies (0–15Hz), where
themajority of the power of the PPG signal is located, the
first-difference differentiator and SNRD frequency responses
arepractically equal.
In practice, biosignals such as PPG, which are related toheart
activity, are recurring but not periodic.This means thatthe
harmonic components of the two consecutive recurrencesof the PPG
signal and its derivatives can be at differentfrequencies. In this
study, the low-pass filter is usedwith static
-
4 The Scientific World Journal
Raw PPG signal
Two-times differentiation with SNRD
Four-times differentiation with SNRD
Signal resampling forselected recurrence
PM low-pass filter
Loop
is ca
rrie
d ou
tfo
r eve
ry re
curr
ence
Averaging of recurrencesAveraging of recurrences Averaging of
recurrences
Signal resampling forselected recurrence
Signal resampling forselected recurrence
PM low-pass filterPM low-pass filter
Loop
is ca
rrie
d ou
tfo
r eve
ry re
curr
ence
Loop
is ca
rrie
d ou
tfo
r eve
ry re
curr
ence
Alignment of recurrences by using 50%
level of raising front as reference point
Alignment of recurrences
Detection of waves: ‘‘a’’, ‘‘b’’, ‘‘c’’, ‘‘d’’, and ‘‘e’’
Detection of zero crossing points
Alignment of recurrences
FIR high-pass filterEdge frequency, 0.5 Hz
FIR low-pass filterEdge frequency, 30 Hz
Figure 3: Block diagram of the signal processing for the second
derivative analysis.
edge frequency. Accordingly, certain numbers of
harmoniccomponents are passed and all others are suppressed.
Thelengths of the PPG signal recurrences are then normalizedto
ensure that all the harmonic components are processed inthe same
way (Figure 3).
The PPG signal is resampled in such a way that one ofthe
selected recurrence lengths is 1 s, which corresponds tothe pulse
frequency of 1Hz. In this case, the fundamentalfrequency is
situated at 1Hz. All the other components lay atthe frequency
multiples of the fundamental frequency. In thenext step, the signal
is filtered with the 1Hz wide transition-band PM filter [19]. The
maximum allowable errors for thepass and stop bands are set at
0.001. The resampling andfiltering are also carried out with the
second and fourthderivatives of the PPG signal. After filtering,
the copy ofselected recurrence is aligned with other normalized
andfiltered recurrences from this PPG signal. The recurrences ofPPG
signals can be aligned by using different distinct pointsfrom the
signal as reference, for example, the recurrencemaximal orminimal
point of the raising front.The recurrenceminimal point can be
difficult to determine, because of thewavy ending of the diastole
phase. It is also difficult todetermine the PPG signal maximal
point as it depends on thestate of the cardiovascular system [20].
The 50 percent levelof the PPG signal raising front is used as the
reference pointfor the alignment of the recurrences. Furthermore,
the secondand fourth derivatives are moved according to the
movementof the PPG signal recurrences.
The resampling, filtering, and aligning processes
outlinedpreviously are carried out separately for every recurrence
inPPG signals. The averaged waveform for one subject with its9
recurrences is given in Figure 4.
Subsequently, the peaks of waves “a”, “b”, “c”, “d”, and “e”are
found from the averaged SDPPG waveform. Firstly, the
zero crossings of the averaged fourth derivative waveformare
found. The peaks of waves “a”, “b”, “c”, “d”, and “e” arebetween
zero crossings of the fourth derivative waveform asit is revealed
in Figure 4. Secondly, the minimal andmaximalpoints of the SDPPG
waveform are located between the zerocrossings of the fourth
derivative waveform. There can bewaveforms of the SDPPG, where the
peaks of the “c” and“d” waves do not appear. In this case, the “c”
and “d” wavesare determined in the places of the SDPPG waveform,
wherethe fourth derivative is maximal or minimal between
zerocrossings.
2.2. Optimization of PM Low-Pass Filter Edge Frequency.The
recurrences and averaged waveform of the SDPPG areaffected by the
edge frequency of the PM low-pass filter.The optimal edge frequency
of the PM low-pass filter wasoptimized in order to achieve the
lowest standard deviationof the SDPPG wave amplitudes, which
ultimately minimizesthe standard deviation of AGI. In addition, the
variation inthe placement of waves “a”, “b”, “c”, “d”, and “e” on
time domainhas to be minimal throughout all the periods for one
subject.Here, it is assumed that the cardiovascular system does
notchange over short periods in cases of healthy subjects.
Theoptimization of the PM low-pass filter edge frequency wascarried
out on signals from a group of healthy subjects.
The width of the PM low-pass filter transition-band was1Hz and
the edge frequency was changed between 4 and14Hz with a step of
1Hz. The order of the correspondingPM filter was 3255 at sampling
frequency 1 kHz. The 3–13 harmonic components in addition to the
fundamentalharmonic are passed through the filter as the
recurrences ofthe SDPPG were normalized to the frequency of 1Hz. In
thisway, the influence of each harmonic component to waves “a”,“b”,
“c”, “d”, and “e” can be analyzed.
-
The Scientific World Journal 5
50% level
Zero crossings
a
b c d e
50% level
b c d e
Zero crossings
PPG signal
2nd derivative of PPG signal (SDPPG)
4th derivative of PPG signal
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6Time (s)
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6Time (s)
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6Time (s)
100
50
0
−50
−100
−150
(a.u
.)(a
.u.)
(a.u
.)
0.03
0.02
0.01
0
−0.01
2
1
0
−1
−2
×10−5
Figure 4: The averaged PPG and its second derivative and
fourthderivative waveforms (black bold line) with filtered and
normalizedrecurrences (thin lines). The recurrences are aligned
according to50% of the PPG signal raising front and the distinct
waves “a”, “b”,“c”, “d”, and “e” are detected by using the zero
crossings of the fourthderivative.
The standard deviations for the normalized amplitudes,b/a, c/a,
d/a, e/a, and AGI were calculated for each edgefrequency. For the
standard deviation calculation, the nor-malized amplitudes,𝑥a,
fromnormalized SDPPGrecurrencesand from the averaged SDPPG waveform
were used. Thenormalized amplitudes from the averaged SDPPG
waveformwere taken as average values 𝑥avg. The standard
deviationswere calculated for signals from each healthy subject, k,
byusing following equation:
𝑆𝐷 (𝑘) =√∑𝑛
𝑖=1(𝑥a (𝑖) − 𝑥avg)
2
𝑛 − 1,
(1)
where 𝑖 is the number of period and 𝑛 is the total numberof
periods in the processed signal. Similarly, the standarddeviation
of the detected wave peaks on the time domain wascalculated. The
standard deviations were averaged over thegroup of subjects by
using following equation:
𝑆𝐷avg =1
𝑚⋅
𝑚
∑𝑘=1
𝑆𝐷 (𝑘) , (2)
where𝑚 is the total number of healthy subjects.
2.3. Pilot Study on Patients. The improved SDPPG signalanalysis
algorithm was tested on the signals from a group ofhealthy and
diabetes patients.The optimal PM low-pass filteredge frequency was
used for the analysis. The SDPPG waveswere detected and AGI values
were calculated with standarddeviations.
2.4. Subjects. The study was performed after approval ofthe
protocol by the Tallinn Ethics Committee on MedicalResearch at the
National Institute for Health Development,Estonia.ThePPG signals
for the analysis were registered fromhealthy subjects and diabetes
patients.
All the subjects in the healthy group were aged from 21to 66
years. They were not on permanent medication andthey were dealing
with various levels of physical activity intheir everyday lives. As
the waveform of the PPG signal varieswith age, the subjects were
divided into the following agegroups: 20–30, 30–40, 40–50, 50–60,
and 60–70. Each agegroup, except the groups of 50–60 and 60–70,
comprised fivesubjects. Those age groups comprised three subjects,
becauseit was difficult to find healthy subjects to fulfill our
criteria. Inall, 21 healthy subjects (𝑚 = 21) participated in the
study.
All subjects in the group of diabetes patients had
receiveddiagnosis from a medical doctor. In all, 20 diabetes
patientsparticipated in this pilot study. The patients were
agedbetween 27 and 66 years. The diabetes patients may
haveincreased arterial stiffness due to the sclerotic processes in
thevessels.
2.5. Instrumentation. All signal processing was carried outin
MATLAB. The high- and low-pass filter coefficients werecalculated
by using the “fir1” function. A separate functionwas written for
calculating coefficients of the SNRD [18].The coefficients of PM
filter were calculated using functions“firpmord” and “firpm.”
The PPG signals were registered from the index fingerby using an
experimental measurement complex [21], whichincluded a Nellcor
finger clip sensor and lab-built PPGmodule, among other devices.
The PPG signal was digitizedwith a PCIMIO-16-1 data acquisition
card and registeredwithLabVIEW environment. The sampling frequency
was 1 kHz.The 1-minute long signal was recorded, and a 15-period
longsegment (𝑛 = 15) was selected for the SDPPG analysis. Thesignal
registration was carried out, while the subject was in aresting
position. The subject was in a resting position at least10minutes
prior to themeasurements.The room temperaturewas around 23 degrees
during the experiments.
3. Results
The general change in standard deviation of the
normalizedamplitudes and AGI in cases of different edge
frequenciesis illustrated in Figures 5(a)–5(e). For each edge
frequency,the given standard deviation is averaged over the group
ofhealthy subjects. The minimal average standard deviationfor AGI,
b/a, c/a, d/a, and e/a is 0.06, 0.02, 0.04, 0.03, and0.02
respectively. For all parameters, the minimal standard
-
6 The Scientific World Journal
0.5
0.4
0.3
0.2
0.1
04 5 6 7 8 9 10 11 12 13 14
Edge frequency of PM filter (Hz)
SDav
gof
AGI
(a)
0.5
0.4
0.3
0.2
0.1
04 5 6 7 8 9 10 11 12 13 14
Edge frequency of PM filter (Hz)
SDav
gof
b/a
(b)
0.5
0.4
0.3
0.2
0.1
04 5 6 7 8 9 10 11 12 13 14
Edge frequency of PM filter (Hz)
SDav
gof
c/a
(c)
0.5
0.4
0.3
0.2
0.1
04 5 6 7 8 9 10 11 12 13 14
Edge frequency of PM filter (Hz)SD
avg
of d/a
(d)
0.5
0.4
0.3
0.2
0.1
04 5 6 7 8 9 10 11 12 13 14
Edge frequency of PM filter (Hz)
SDav
gof
e/a
(e)Figure 5: The average standard deviations (𝑆𝐷avg) of AGI and
normalized amplitudes at different edge frequencies for 21 healthy
subjects.(a) AGIs at different edge frequencies; (b) normalized
amplitudes b/a at different edge frequencies; (c) normalized
amplitudes c/a at differentedge frequencies; (d) normalized
amplitudes d/a at different edge frequencies; (e) normalized
amplitudes e/a at different edge frequencies.
deviationwas foundwhere the edge frequency of the PMfilteris 6Hz
and transition band is 1Hz.
Similarly, in Figures 6(a)–6(e), the standard deviationsto
characterize the dispersion of wave peaks “a”, “b”, “c”, “d”,and
“e” in the time domain are given. The given standarddeviations are
averaged over the group of healthy subjects.The minimal average
standard deviations for wave peaks “a”,“b”, “c”, “d”, and “e” in
the time domain are 2.2ms, 1.9ms,4.6ms, 2.8ms, and 5.0ms,
respectively.Theminimal standarddeviations can be found for the
edge frequency of 6Hz for allwaves, except for wave “b”. In the
case of wave “b”, theminimalstandard deviation was at edge
frequency of 4Hz.
For the purpose of comparison, the same PPG signalswere also
processed with the algorithm described by Mil-lasseau et al. [17].
The average standard deviation for AGIvalue is 0.12.
The AGI and age relationship for the healthy subjects
anddiabetes patients with standard deviation bars are illustratedin
Figure 7(a). The PM filter edge frequency and transition-band were
6Hz and 1Hz, respectively, which according tothe previously
presented results seems to be optimal for
the SDPPG analysis. In addition, regression analysis wascarried
out in order to estimate the relationship betweenAGI and age by
using generalized linear model. As a firstapproach the general
linear model was used, which is a caseof the generalized linear
model with identity link function.A following regression model was
proposed: 𝑦AGI = 0.019x −1.556. Despite of relatively simple model
high correlation𝑟 = 0.91 was found between AGI and age for the
healthygroup, which shows the strong linear relationship betweentwo
variables.
In Figure 7(b) it can be seen the Bland-Altman plot forthe
proposed model 𝑦AGI. The standard deviations for themodel a 𝑆𝐷AGI =
0.126. For diabetes patients the mean AGIvalue difference from the
proposed model 𝑦AGI is 𝑚𝑒𝑎𝑛Dia=0.359. The AGI differences from the
proposed model 𝑦AGIwere compared betweenhealthy anddiabetes
patients groups.Paired 𝑡-test (two-sample assuming unequal
variances) wasperformed in MS Excel with 𝛼 = 0.05. The significance
levelof paired 𝑡-test was 𝑃 < 0.0005, which shows the
differencebetween two groups.
-
The Scientific World Journal 7
0123456789
10
4 5 6 7 8 9 10 11 12 13 14Edge frequency (Hz)
SDav
gof
wav
e ‘‘a”
(ms)
(a)
05
1015202530
4 5 6 7 8 9 10 11 12 13 14Edge frequency (Hz)
SDav
gof
wav
e ‘‘b”
(ms)
(b)
0102030405060708090
100
4 5 6 7 8 9 10 11 12 13 14Edge frequency (Hz)
SDav
gof
wav
e ‘‘c”
(ms)
(c)
0102030405060708090
100
4 5 6 7 8 9 10 11 12 13 14Edge frequency (Hz)
SDav
gof
wav
e ‘‘d”
(ms)
(d)
0102030405060708090
100
4 5 6 7 8 9 10 11 12 13 14Edge frequency (Hz)
SDav
gof
wav
e ‘‘e”
(ms)
(e)
Figure 6: The average standard deviations (𝑆𝐷avg) of the waves
“a”, “b”, “c”, “d”, and “e” on the time domain at different edge
frequenciesfor 21 healthy subjects. (a) standard deviation of wave
“a” at different edge frequencies; (b) standard deviation of wave
“b” at different edgefrequencies; (c) standard deviation of wave
“c” at different edge frequencies; (d) standard deviation of wave
“d” at different edge frequencies;(e) standard deviation of wave
“e” at different edge frequencies.
0
0.4
0.8
1.2
10 20 30 40 50 60 70
AGI (
a.u.)
Age (years)
Age group 20–30Age group 30–40Age group 40–50
Age group 50–60Age group 60–70Diabetes patients
−0.4
−0.8
−1.2
−1.6
r = 0.91yAGI = 0.019x − 1.556
(a)
00.20.40.60.8
11.2
10 20 30 40 50 60 70Age (years)
−0.2
−0.4
MeanDia = 0.359
SDAGI = +0.126
SDAGI = −0.126
AGI−y
AGI
(a.u
.)
(b)
Figure 7: (a) The AGI data points with standard deviation (SD)
bars for groups of healthy subjects and diabetes patients. The
linear modelline 𝑦AGI is constructed for a group of healthy
subjects. (b) Bland-Altman plot for constructed model. With dotted
line the standard deviationlevels for group of healthy subjects are
given. With dashed line is given the mean AGI difference from
linear model for diabetes patients.
-
8 The Scientific World Journal
4. Discussions
With an improved SDPPG analysis algorithm, the averagestandard
deviation for the AGI value is 0.06, which consti-tutes about 5% of
the whole scale of AGI [12]. Compared tothe algorithm of Millasseau
et al. [17], the average standarddeviation is twice lower. As a
result, subjects with increasedarterial stiffness can be more
easily differentiated fromhealthy subjects, and the prevention of
cardiovascular diseasecan be improved.
The relatively high correlation relation was foundbetween AGI
and age by using the algorithm with optimaledge frequency (Figure
7(a)). This is in relation to previouslypublished results by
Takazawa et al. [12], in which a goodcorrelation betweenAGI and age
among healthy subjects wasshown. There are still some deviations
from the regressionmodel line, 𝑦AGI, which can be caused by the
impact ofcardiovascular deficiencies and the subject’s biological
age.In addition model can be more complex and dependent
onadditional variables, such as blood pressure. However, thisshould
be considered in the scope of future studies.
The noticeably higher AGI values, compared to thehealthy group
of subjects, were found for diabetes patients(Figure 7(a)).The same
behavior is also visible in Figure 7(b).Furthermore, the
statistically significant difference was foundbetween the healthy
subjects and diabetes patients. Thehigher AGI values are caused by
the increased arterialstiffness of diabetes patients. Nevertheless,
some of the dia-betes patients have similar AGI values compared to
healthysubjects. It can be caused by the early diagnosis of
diabetesmellitus, which is followed with efficient therapy, and as
aresult premature stiffening of the arteries has been stopped.
It can be seen from Figures 5(a)–5(e) that the lowestaverage
standard deviation was achieved when the edgefrequency of PM filter
is 6Hz. Close to 6Hz, the normalizedrecurrences start to resemble.
The larger standard deviationson higher edge frequencies are caused
by the noise andunwanted frequency components of the PPG signal,
whichare situated at higher frequencies and amplified
throughdifferentiation. This causes the faulty detection of the
wavesfrom the single normalized recurrence and averaged
SDPPGwaveform. At lower edge frequencies, the harmonic com-ponents
are suppressed, which form waves “a”, “b”, “c”, “d”,and “e”, and
the peaks of waves, “c” and “d”, were missingin single normalized
recurrences. As a result, the amplitudeof the waves in single
normalized recurrences and averagedSDPPG waveform are different,
which caused the increasein standard deviation (Figures 5(a)–5(e)).
This means that itis necessary to have a fundamental harmonic with
5 higherharmonic components in order to detect waves “a”, “b”, “c”,
“d”,and “e” from the PPG signal.
The dispersion of wave peaks on time domain decreases,similar to
the results seen in Figure 5, when the edge fre-quency of PM filter
approaches 6Hz (Figures 5(a)–6(e)). Asin Figure 5, the detection of
thewaves from the single normal-ized recurrences and from the
averaged SDPPG waveformcan be different at higher frequencies,
which increases thestandard deviation. At frequencies lower than
6Hz, the wavepeaks can bemissing in single normalized recurrence
and the
detection point is shifted comparedwith the averaged
SDPPGwaveform.
5. Conclusions
In conclusion, it can be said that the standard deviationof AGI
values is minimized by using the improved SDPPGalgorithm.
Furthermore, the diabetes patients have noticeablyhigher AGI
values, which are caused by an increase in thearterial stiffness.
As a result, the subjects, with increasedarterial stiffness can be
more easily differentiated fromhealthy subjects and the prevention
of cardiovascular diseasecan be improved. As a future study the
more complex modelshould be considered in order to enhance the
discriminationof the healthy subjects and patients with increased
stiffnessby taking into account additional physiological
variables.In addition proposed algorithm should be compared
withsimilar arterial stiffness estimation reference methods.
Acknowledgments
Thiswork was supported by the Estonian Science FoundationGrant
no. 7506, by the Estonian Targeted Financing ProjectSF0140027s07,
and by the European Union through theEuropean Regional Development
Fund.
References
[1] M. W. Rajzer, W. Wojciechowska, M. Klocek, I. Palka,
M.Brzozowska-Kiszka, and K. Kawecka-Jaszcz, “Comparison ofaortic
pulse wave velocity measured by three techniques: com-plior,
sphygmocor and arteriograph,” Journal of Hypertension,vol. 26, no.
10, pp. 2001–2007, 2008.
[2] S. S. Silvera, H. E. Aidi, J. H. F. Rudd et al.,
“Multimodalityimaging of atherosclerotic plaque activity and
compositionusing FDG-PET/CT and MRI in carotid and femoral
arteries,”Atherosclerosis, vol. 207, no. 1, pp. 139–143, 2009.
[3] M. R. Skilton, L. Boussel, F. Bonnet et al., “Carotid
intima-media and adventitial thickening: comparison of new and
estab-lished ultrasound andmagnetic resonance imaging
techniques,”Atherosclerosis, vol. 215, no. 2, pp. 405–410,
2011.
[4] C. F. Clarenbach, A. Stoewhas, A. J. R. Van Gestel et al.,
“Com-parison of photoplethysmographic and arterial
tonometry-derived indices of arterial stiffness,”Hypertension
Research, vol.35, no. 2, pp. 228–233, 2012.
[5] J. Allen, “Photoplethysmography and its application in
clinicalphysiological measurement,” Physiological Measurement,
vol.28, pp. R1–R39, 2007.
[6] A. A. R. Kamal, J. B. Harness, G. Irving, and A. J.
Mearns,“Skin photoplethysmography—a review,” Computer Methodsand
Programs in Biomedicine, vol. 28, no. 4, pp. 257–269, 1989.
[7] L. G. Lindberg and P. Å. Öberg, “Optical properties of
blood inmotion,” Optical Engineering, vol. 32, pp. 253–257,
1993.
[8] K. Pilt, K. Meigas, R. Ferenets, and J. Kaik,
“Photoplethysmo-graphic signal processing using adaptive sum comb
filter forpulse delay measurement,” Estonian Journal of
Engineering, vol.16, no. 1, pp. 78–94, 2010.
-
The Scientific World Journal 9
[9] M. Huotari, N. Yliaska, V. Lantto, K. Määttä, and J.
Kosta-movaara, “Aortic and arterial stiffness determination by
pho-toplethysmographic technique,” Procedia Chemistry, vol. 1,
no.1, pp. 1243–1246, 2009.
[10] S. C. Millasseau, J. M. Ritter, K. Takazawa, and P.
J.Chowienczyk, “Contour analysis of the photoplethysmographicpulse
measured at the finger,” Journal of Hypertension, vol. 24,no. 8,
pp. 1449–1456, 2006.
[11] U. Rubins, “Finger and ear photoplethysmogram
waveformanalysis by fitting with Gaussians,” Medical and
BiologicalEngineering and Computing, vol. 46, no. 12, pp.
1271–1276, 2008.
[12] K. Takazawa, N. Tanaka, M. Fujita et al., “Assessment
ofvasoactive agents and vascular aging by the second derivativeof
photoplethysmogram waveform,” Hypertension, vol. 32, no.2, pp.
365–370, 1998.
[13] J. Hashimoto, D. Watabe, A. Kimura et al., “Determinantsof
the second derivative of the finger photoplethysmogramand
brachial-ankle pulse-wave velocity: The Ohasama study,”American
Journal of Hypertension, vol. 18, no. 4, pp. 477–485,2005.
[14] T. Otsuka, T. Kawada, M. Katsumata, and C. Ibuki, “Utility
ofsecond derivative of the finger photoplethysmogram for
theestimation of the risk of coronary heart disease in the
generalpopulation,” Circulation Journal, vol. 70, no. 3, pp.
304–310,2006.
[15] T. Otsuka, T. Kawada, M. Katsumata, C. Ibuki, and Y.
Kusama,“Independent determinants of second derivative of the
fingerphotoplethysmogram among various cardiovascular risk fac-tors
in middle-aged men,” Hypertension Research, vol. 30, no.12, pp.
1211–1218, 2007.
[16] A. Grabovskis, Z. Marcinkevics, Z. Lukstina et al.,
“Usability ofphotoplethysmography method in estimation of conduit
arterystiffness,” inNovel Biophotonic Techniques and Applications,
vol.8090 of Proceedings of SPIE, May 2011.
[17] S. C. Millasseau, R. P. Kelly, J. M. Ritter, and P. J.
Chowienczyk,“The vascular impact of aging and vasoactive drugs:
compar-ison of two digital volume pulse measurements,”
AmericanJournal of Hypertension, vol. 16, no. 6, pp. 467–472,
2003.
[18] P. Holoborodko, “Smooth noise robust differentiators,”
http://www.holoborodko.com/pavel/numerical-methods/numerical-derivative/smooth-low-noise-differentiators/,
2008.
[19] T. W. Parks and J. H. McClellan, “Chebyshev approximation
fornonrecursive digital filters with linear phase,”
IEEETransactionsCircuit Theory, vol. 19, no. 2, pp. 189–194,
1972.
[20] W. W. Nichols and M. F. O’Rourke, McDonald’s Blood Flowin
Arteries: Theoretic, Experimental and Clinical
Principles,Hodder-Arnold, London, UK, 5th edition, 2005.
[21] K. Pilt, K. Meigas, M. Viigimaa, K. Temitski, and J. Kaik,
“Anexperimental measurement complex for probable estimationof
arterial stiffness,” in Proceedings of the 32nd
InternationalConference of the IEEE Engineering in Medicine and
BiologySociety, pp. 194–197, Buenos Aires, Argentina, 2010.
-
Submit your manuscripts athttp://www.hindawi.com
VLSI Design
Hindawi Publishing Corporationhttp://www.hindawi.com Volume
2014
International Journal of
RotatingMachinery
Hindawi Publishing Corporationhttp://www.hindawi.com Volume
2014
Hindawi Publishing Corporation http://www.hindawi.com
Journal ofEngineeringVolume 2014
Hindawi Publishing Corporationhttp://www.hindawi.com Volume
2014
Shock and Vibration
Hindawi Publishing Corporationhttp://www.hindawi.com Volume
2014
Mechanical Engineering
Advances in
Hindawi Publishing Corporationhttp://www.hindawi.com Volume
2014
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawi Publishing Corporationhttp://www.hindawi.com Volume
2014
Hindawi Publishing Corporationhttp://www.hindawi.com Volume
2014
Electrical and Computer Engineering
Journal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume
2014
Distributed Sensor Networks
International Journal of
The Scientific World JournalHindawi Publishing Corporation
http://www.hindawi.com Volume 2014
SensorsJournal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume
2014
Modelling & Simulation in EngineeringHindawi Publishing
Corporation http://www.hindawi.com Volume 2014
Hindawi Publishing Corporationhttp://www.hindawi.com Volume
2014
Active and Passive Electronic Components
Hindawi Publishing Corporationhttp://www.hindawi.com Volume
2014
Chemical EngineeringInternational Journal of
Control Scienceand Engineering
Journal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume
2014
Antennas andPropagation
International Journal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume
2014
Hindawi Publishing Corporationhttp://www.hindawi.com Volume
2014
Navigation and Observation
International Journal of
Advances inOptoElectronics
Hindawi Publishing Corporation http://www.hindawi.com
Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume
2014
New Photoplethysmographic - TP169035